Download XML-Relational Mapping: Techniques for XML Data in Databases and more Slides Database Management Systems (DBMS) in PDF only on Docsity!
XML-Relational Mapping
CPS 216
Advanced Database Systems
2
Announcements (March 18)
Midterm sample solution available outside my office
Course project milestone 2 due March 30
Homework #3 due April 6
Talk by Amol Deshpande
Adaptive Query Processing to Handle Estimation Errors
Monday, 11:30am-12:30pm, D
Reading assignment due next Monday
Two VLDB papers on native XML databases
3
Approaches to XML processing
Text files (!)
Specialized XML DBMS
Lore (Stanford), Strudel (AT&T), Tamino/QuiP
(Software AG), X-Hive, Timber (Michigan), etc.
Still a long way to go
Object-oriented DBMS
eXcelon (ObjectStore), ozone, etc.
Not as mature as relational DBMS
Relational (and object-relational) DBMS
Middleware and/or object-relational extensions
Mapping XML to relational
Store XML in a CLOB (Character Large OBject) column
Simple, compact Full-text indexing can help (often provided by DBMS vendors as object-relational “extensions”)
Alternatives?
Schema-oblivious mapping: well-formed XML → generic relational schema
- Node/edge-based mapping for graphs
- Interval-based mapping for trees
- Path-based mapping for trees Schema-aware mapping: valid XML → special relational schema based on DTD
5
Node/edge-based: schema
Element ( eid , tag )
Attribute ( eid , attrName , attrValue )
Attribute order does not matter
ElementChild ( eid , pos , child )
pos specifies the ordering of children child references either Element ( eid ) or Text ( tid )
Text ( tid , value )
tid cannot be the same as any eid
) Need to “invent” lots of id ’s
) Need indexes for efficiency, e.g., Element ( tag ), Text ( value )
6
Node/edge-based: example
Foundations of Databases Abiteboul Hull Vianu Addison Wesley 1995 …
eid tag e0 bibliography e1 book e2 title e3 author e4 author e5 author e6 publisher e7 year
Element
eid pos child e0 1 e e1 1 e e1 2 e e1 3 e e1 4 e e1 5 e e1 6 e e2 1 t e3 1 t e4 1 t e5 1 t e6 1 t e7 1 t
ElementChild
eid attrName attrValue e1 ISBN ISBN- e1 price 80
Attribute
tid value t0 Foundations of Databases t1 Abiteboul t2 Hull t3 Vianu t4 Addison Wesley t5 1995
Text
Interval-based: schema
Element ( left , right , level , tag )
left is the start position of the element right is the end position of the element level is the nesting depth of the element (strictly speaking, unnecessary) Key is left
Attribute ( left , attrName , attrValue )
Text ( left , level , value )
) Where did ElementChild go?
11
Interval-based: example
1 2 34Foundations of Databases 67Abiteboul 910Hull 1213Vianu 1516Addison Wesley 18191995 21… 999 (^) bibliography
book
title author author author publisher year
1,999,
2,21,
3,5,3 6,8,3 9,11,3 12,14,3 15,17,3 18,20,
12
Interval-based: queries
//section/title
SELECT e2.left FROM Element e1, Element e WHERE e1.tag = ‘section’ AND e2.tag = ‘title’ AND e1.left < e2.left AND e2.right < e1.right AND e1.level = e2.level-1;
)Path expression becomes “containment” joins!
- Number of joins is proportional to path expression length
//book//title
SELECT e2.left FROM Element e1, Element e WHERE e1.tag = ‘book’ AND e2.tag = ‘section’ AND e1.left < e2.left AND e2.right < e1.right;
)No recursion!
How about XQuery?
DeHaan et al. SIGMOD 2003
Evaluating an XQuery expression results in a sequence of
environments
An environment E maps each query variable v to its value: a forest of XML trees (a node-set) fv
Encode using tables with “dynamic intervals”
Table I : increasing sequence of integers, one per environment For each query variable v , create a table Tv ( s (tring) , l (eft) , r (ight) ) representing the value of v in all environments
- Sorted on l to support efficient processing
- Different environments form non-overlapping regions
14
Example T v
15
Translating /
Given Tv for values of v , compute v /name
A path-based mapping
Label-path encoding
Element ( pathid , left , right , value ), Path ( pathid , path )
path is a label path starting from the root
Why are left and right still needed?
pathid left right … 1 1 999 … 2 2 21 … 3 3 5 … 4 6 8 … 4 9 11 … 4 12 14 … … … … …
Element
pathid path 1 /bibliography 2 /bibliography/book 3 /bibliography/book/title 4 /bibliography/book/author … …
Path
20
Label-path encoding: queries
Simple path expressions with no conditions
//book//title
Perform string matching on Path Join qualified pathid ’s with Element
Path expression with attached conditions need to be broken
down, processed separately, and joined back
//book[publisher=‘Prentice Hall’]/title
Evaluate //book Evaluate //book/title Evaluate //book/publisher[text()=‘Prentice Hall’] Join to ensure title and publisher belong to the same book
21
Another path-based mapping
Dewey-order encoding
Each component of the id represents the order of the
child within its parent
Unlike label-path, this encoding is “lossless”
bibliography
book
title author author author publisher year
1
1.1.1 1.1.2 1.1.3 1.1.4 1.1.5 1.1.
Dewey-order encoding: queries
Examples:
//title
//section/title
//book//title
//book[publisher=‘Prentice Hall’]/title
23
Schema-aware mapping
Idea: use DTD to design a better schema
Basic approach: elements of the same type go into one table
Tag name → table name Attributes → columns
- If one exists, ID attribute → key column; otherwise, need to “invent” a key
- IDREF attribute → foreign key column Children of the element → foreign key columns
- Ordering of columns encodes ordering of children
… ]>
book ( ISBN , price , title_id , …) title ( id , PCDATA_id ) PCDATA ( id , value )
24
Handling * and + in DTD
What if an element can have any number of children?
Example: Book can have multiple authors
book ( ISBN , price , title_id , author_id , publisher_id , year_id )? )BCNF?
Idea: create another table to track such relationships
book ( ISBN , price , title_id , publisher_id , year_id ) book_author ( ISBN , author_id ) )BCNF decomposition in action! )A further optimization: merge book_author into author
Need to add position information if ordering is important
book_author ( ISBN , author_pos , author_id )
Pros and cons of inlining
Not always applicable
29
Result restructuring
Simple results are fine
Each tuple returned by SQL gets converted to an element
Simple grouping is fine (e.g., books with multiple authors)
Tuples can be returned by SQL in sorted order; adjacent tuples are grouped into an element
Complex results are problematic: one SQL query only
returns a single table; columns cannot contains sets or
structures
E.g., books with multiple authors and multiple references
- Option 1: one table with all combo of authors/references → bad
- Option 2: two tables, one w/ authors and the other w/ references → join is done as post processing
- Option 3: sorted “union” of NULL-padded authors and references
30
Comparison of approaches
Schema-oblivious
Flexible and adaptable; no DTD needed Queries are easy to formulate
- Translation from Xpath/XQuery can be easily automated Queries involve lots of join and are expensive
Schema-aware
Less flexible and adaptable Need to know DTD to design the relational schema Query formulation requires knowing DTD and schema Queries are more efficient XQuery is tougher to formulate because of result restructuring
